They are asking the Bitcoin community - what information do you want to try to extract from the transaction graph?
Blockchain-based provably fair gambling sites like SatoshiDICE use a publicly known address for each wager and for each payout. This causes them to be good candidates for use in study.
I'm not sure what information it might yield. Since individual wagers there frequently are in the 100 to 250 BTC range (i.e., $1,200 to $3,000 USD for a single bet) it would be interesting to know if somehow this service is being used as a
low cost mixing service.
While this doesn't remove traceability, it does change the composition of a wallet. So taking coins that are tainted and sending the coins through SatoshiDICE is something that costs 1.9% per pass but it cuts the level of original taint by 50% (statistically, over the long run). Of course, this is a flawed approach for mixing because the winnings returned would assume the same level of taint that the coins used in the wagering (i.e., a tainted coin taints every other coin in a SatoshiDICE payout) had. But the tools that report taint probably aren't sophisticated enough to take this characteristic of SatoshiDICE payouts into account and thus report low levels of taint after being passed through SatoshiDICE.
I don't know how you could identify wagers made for the purpose of mixing (reducing taint) versus wagers made for gambling / entertainment purposes though.
